7
No mention of how it compares to existing spatial indexing methods such as R[*]-trees. That was my first thought reading the article, but they only give a comparison to naïve NxM testing. I assume this method is still an improvement in the presence of sharding, but doubt it's the 400× quoted.